Long videos with full control - Wan 2.2 I2V (SVI 2 Pro). Individual LoRa, multiple reference images and more — v2.0

Use this workflow to create long, segmented videos from prompts and reference images with Wan Video 2.2 I2V-A14B; set each section’s prompt, length, seed, base model, and LoRa, then produce section images, saved video data, and a combined video with separate upscaling or frame-­­

Add sections in one continuous workflow to extend a video, then load saved video data to continue from where you stopped, including after a ComfyUI restart.

For each section, provide a reference image, prompt, length, and seed; select a base model and LoRa when needed.

Outputs include section images and previews, saved Wan video latents, a stitched full video, upscaling, and RIFE frame interpolation; the v2.0 notes say the last image of each section is saved and previews appear immediately after generation.

Use the guide inside the Options subgraph as your starting point.

If you are following the v3.0 notes, they name ComfyUI-HeresyNodes and say old ComfyUI-WanVideo-LatentIO nodes used in v2 can be removed if they cause conflicts.

The setup list names these model files: RealESRGAN_x2plus.pth, rife49.pth, umt5_xxl_fp8_e4m3fn_scaled.safetensors, wan_2.1_vae.safetensors, wan_latent.safetensors, wan22_smoothMix_V20_HIGH.safetensors, and wan22_smoothMix_V20_LOW.safetensors.

The setup also names these node packs: basic_data_handling, comfy-image-saver, comfy-mtb, comfyui_essentials, ComfyUI_JPS-Nodes, ComfyUI-Crystools, comfyui-easy-use, and comfyui-frame-interpolation.

Additional node packs named are comfyui-kjnodes, comfyui-various, comfyui-videohelpersuite, comfyui-vslinx-nodes, mikey_nodes, and rgthree-comfy.

Choose a different base model and LoRa for each section.

Use a different reference image for each section to introduce assets or characters that do not appear in the starting image; the description says this does not require a LoRa.

Run upscaling and frame interpolation separately from generation, stitch a full video from saved data, or generate through those steps in one pass.

The listed version is v2.0, but the description also contains v3.0 setup notes; verify which instructions match the copy you have.

Suggestion · non vérifié

Before running, verify that your ComfyUI setup contains the exact model filenames and node-pack names listed here.

Suggestion · non vérifié